JP3117091B2 - Image coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for encoding image information.

[0002]

2. Description of the Related Art For example, in a facsimile machine,
Images to be transmitted or received images are stored in the page memory. In the image editing apparatus, an image to be edited or an image after editing is also stored in the page memory. The required capacity of the page memory increases as the image size increases and as the resolution increases. Further, when storing a color image, a capacity three times as large as that of a single color image is required. For example, when storing an A3-size full-color image at a resolution of 400 dpi (dots / inch), the memory capacity is 128M.
Even bytes. When the memory capacity is increased in this way, there is a problem that the price of the page memory becomes high, and that reading / writing of image data takes a long time and a processing time becomes long.

As a solution to such a problem, an image signal is highly efficiently encoded in a form that can be edited as encoded data,
Reduction of page memory capacity is being considered. Such encoding requires the following three characteristics.

[0004] First, the compression ratio is uniform.
That is, since the page memory has a finite capacity, it is necessary that the page memory can be encoded at a predetermined compression ratio without depending on the image. Second, it is possible to edit the encoded data as it is. That is, 2 of the image signal
Since the dimensional position is directly obtained from the encoded data, it is necessary that the image signal divided into a predetermined unit can be encoded with a fixed code amount, and that the encoding / decoding can be independently performed for each predetermined unit. It is. Third, the encoding / decoding process is uniform. That is, since encoding is performed on the page memory, it is necessary that encoding / decoding can be performed at high speed and at a constant speed.

In a conventional image coding apparatus for storage and transmission, the compression rate fluctuates due to the fluctuation of the redundancy for each image signal because it is necessary to suppress the spatial and frequency redundancy of the image signal as much as possible. . In addition, there is a tendency to introduce more advanced encoding processing, and encoding / coding is performed independently for each predetermined image division unit.
It is difficult to perform a decoding process. Furthermore, the introduction of the adaptive processing causes the amount of calculation required for the encoding / decoding processing to fluctuate greatly according to the fluctuation of the redundancy for each image signal.
It was difficult to satisfy the above characteristics.

[0006] One of the conventional encoding methods is disclosed in
As described in Japanese Patent Application Laid-Open No. 7-174984, a BTC (Block Truncation Code) which divides an image signal into blocks of a certain size and approximates the shape of each block.
ng) There is a method called an encoding method.

The outline of the BTC coding method is as follows.
This method, the image shown in FIG. 16, L _i × shown in FIG. 17
It is divided into blocks of _Lj pixels. L = L _i = L _j ,
Assuming that the pixel density in the block is a _ij , the average density P _{0 of the} entire block is P ₀ = Σa _ij / L ² . As shown in FIG. 18, the average density P _{0 of the} entire block
The average density and the number of pixels of smaller density pixels are P ₁ , N
_{If 1} ,

[0008]

(Equation 1)

[0009] Assuming that the average density and the number of pixels having a density higher than P ₀ are P ₂ and N ₂ ,

[0010]

(Equation 2)

## EQU1 ## However, when a _ij ≦ P ₀ , φ _ij =
0, a _ij > P ₀ satisfies the condition φ _ij = 1.

Now, by introducing L ² and integers m and n smaller than the number of gradation levels of density, | P ₁ −P ₂ | <m,
Or N ₁ <n or when the N ₂ <n, the density distribution in the block is judged to uniformly, as all phi _ij 0, representing the entire block in just concentration P ₀ as shown in FIG. 19. | P ₁ −P ₂ | ≧ m and N ₁ ≧ n and N ₂ ≧
In the case of n, it is determined that the density distribution in the block is not uniform, and the block is represented by the densities P ₁ and P ₂ as shown in FIG. In this case, this corresponds to placing P ₁ where φ _ij is 0 and P ₂ where φ _ij is ₁ . φ _ij is called resolution information, and P ₀ or P ₁ and P ₂ are called gradation information. The resolution information is collectively encoded for several lines and encoded by the existing binary encoding method, and the continuous length of the block having the same information value is encoded by the existing run length encoding method and transmitted. The parameters m and n in this encoding method are such that m has a role of a threshold for removing isolated noise of an image, and n has a role of a threshold for removing a slight fluctuation of density in a block, and both m and n are large values. The more the image is averaged.

[0013]

However, in this method, since the pixels in a block can be represented by at most two gradations, the block size L is increased to increase the coding efficiency.
Is large, there is a problem that image quality is greatly deteriorated, particularly in a region where the density gradient is gentle. In addition, since the resolution information is uniformly allocated to all the blocks, there is a disadvantage that the redundancy is high. For this reason, it has been attempted to reduce the redundancy by binarizing the resolution information for several lines at the same time, but this is not sufficient.

Further, there is a problem that it is difficult to control the code amount by selecting the parameters m and n, and it is also difficult to edit the image as it is.

In order to eliminate the above-mentioned drawbacks of the prior art, the present invention has been extended so that a block can be expressed using a plurality of different gray scales depending on the contents of an image, and the blocks corresponding to the used gray scales can be used. By transmitting only the necessary resolution information, it is possible to reduce the image quality even when the coding efficiency is increased by increasing the block size, and to use a coding method with high coding efficiency because no redundant resolution information is transmitted. It is intended to be realized.

Further, according to the present invention, each time an input image signal is divided into blocks, the waveform / gain is analyzed by the characteristic amount analyzing means to determine an encoding mode, and the coding mode is determined by the subsequent block approximating encoding means. It is an object of the present invention to perform a block approximate coding to freely change a combination of coding modes without having to determine a coding mode in advance.

[0017]

According to the present invention, there is provided an image coding apparatus comprising: a sampling unit which samples an image and divides the image into an input block of m × n pixels (m and n are positive integers); Block classification means for classifying the gradation and resolution of the input block; and a plurality of block approximate encodings in which the code amount of the gradation and resolution in the input block is allocated to a predetermined code amount for each of the input blocks. Means for adaptively switching the plurality of block approximate encoding means in accordance with the result of classification by the classification means to perform block approximate encoding.

The image coding apparatus according to the present invention further comprises: a block forming means for sampling an image and dividing the sample into an input block of m × n pixels (m and n are positive integers) comprising a plurality of pixels; Average value calculating means for calculating an average value of the average value, average value separating means for subtracting the average value obtained by the average value calculating means from each pixel in the input block, and average value obtained by the average value separating means. Analysis means for analyzing the feature amount of the separation block in the resolution direction and the gradation direction; and a pixel thinning shape and a pixel thinning ratio in the plurality of average value separation blocks preset from the result of the analysis means; Mode determining means for independently determining the number of tones of pixels in the value separating block, and the average value separating block according to the pixel thinning shape and the pixel thinning ratio determined by the mode determining means. Resolution approximation means for thinning out pixels in the block, and gradation approximation means for quantizing pixels in the average value separation block thinned out by the resolution approximation means with the number of gradations determined by the mode determination means. Multiplexing means for multiplexing the average value from the average value calculation means, the determination result of the mode determination means, and the output of the tone approximation means to form code data , wherein the analysis means
When analyzing the feature amount of the input block in the resolution direction
In the above, m × n pixels (m and n are positive integers) obtained in advance,
Alternatively, pixels divided by the positive integer ratio j (j is a positive integer)
Each of a plurality of sets of representative shape blocks consisting of
Calculate the degree of approximation to the value separation block, and calculate the
Index of table-shaped block, or divided into j
Of the representative shape block with the highest degree of approximation for each block
The set of indices is assigned to the
1, and at least the index or the
Is the ratio of the j sets of indices that match each other
A parameter indicating the complexity of the shape of the mean separation block
And the complexity is defined as the second feature amount in the resolution direction.
It is characterized by doing.

[0019]

According to the present invention, for example, a block is extended using a plurality of gradation numbers from one or two gradations to two or more gradation numbers m so as to correspond to the used gradation numbers. Only necessary resolution information is transmitted. As a result, even when the coding block size is relatively large in order to increase the coding efficiency, the image quality is less deteriorated.
Also, as is generally known by Weber's law, a slight gradation difference is detected in an image region having a flat density gradient, but a slight gradation difference is detected in an image region having a sharp change in density. There is a human visual characteristic that is difficult. Therefore,
According to the present invention, in an image region having a flat density gradient, the resolution is lowered while increasing the number of gradations, and conversely, in an image region where the density change is sharp, the resolution is increased while reducing the number of gradations. This makes it difficult to visually detect deterioration in image quality due to encoding, and also avoids encoding redundant gradation information and resolution information, thereby improving encoding efficiency.

Further, in the present invention, an image is sampled and divided into input blocks of m × n pixels composed of a plurality of pixels by a blocking means. Next, the average value in the divided input block is obtained by the average value calculation means, and the average value is subtracted from each pixel in the input block by the average value separation means to obtain an average value separation block. Next, the analysis means analyzes the feature values of the average value separation block in the resolution direction and the gradation direction. Based on the analysis result, a plurality of preset pixel thinning shapes and pixel thinning ratios and the number of gradations of the pixels are determined by the mode determining means. The pixels in the average value separation block are decimated by the resolution approximating means according to the pixel decimating shape and the pixel decimating ratio decided by the mode deciding means, and then the gradation approximating means is performed by the gradation number decided by the mode deciding means Are quantized. The average value from the average value calculation unit, the determination result of the mode determination unit, and the output of the tone approximation unit are multiplexed by a multiplexing unit to form code data.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The features of the present invention will be described in detail below with reference to the drawings.

FIG. 1 shows a schematic configuration example of an image encoding apparatus according to the present invention.

First, the outline of the operation of the image coding apparatus according to the present embodiment will be described. The digitized image information (input image) 1 to be encoded is converted into m
× n (m and n are integers of 1 or more) are divided into pixels. Blocked image information (input block) 3
Is an encoder 4 that performs encoding with different numbers of gradations
a to 4c. In the example of FIG. 1, the encoder 4a performs one-level block approximation encoding, and
b performs two-tone encoding, and the encoder 4c performs m-tone encoding. Encoded image information 5 from each of encoders 4a to 4c
a to 5c are commonly supplied to the selector 6.

The block determiner 7 determines the degree of the density gradient of the input block 3 and outputs a result 8 of the determination to the selector 6. In this embodiment, the selector 6 selects the encoder 4a when the density change is drastic, and selects the encoder 4c when the density gradient is flat. As will be described later, with respect to the resolution information, encoding is performed such that the encoder 4a that performs block approximation encoding of one gradation is large and the encoder 4c that performs block approximation encoding of n gradations is small. Done. Therefore, when the density change is drastic, encoding is performed with a small amount of gradation information and a large amount of resolution information, and when the density gradient is flat, encoding is performed with a large amount of gradation information and a small amount of resolution information. . The selector 6 selects the encoded image information 5a to 5c encoded by any of the encoders 4a to 4c according to the block determination result 8 and outputs the selected encoded image information 9 as selected encoded image information 9.

Next, the operation of the above-described image coding apparatus will be described in detail.

In this embodiment, the block size is 8 pixels ×
It is assumed that there are eight pixels, and the image information 3 in FIG. 1 is input for each block of 8 pixels × 8 pixels. Also, 2 gradations, 4
It is assumed that three types of block approximation encoding of gradation and 16 gradations are performed. However, the block size may be another value, and the block shape is not limited to a square.

As described above, the block determining unit 7 shown in FIG. 1 determines whether the density gradient of the pixels in the block is flat, the density changes drastically, or the density distribution is intermediate. I do. Specifically, the variance of the pixel density in the block is calculated, and if the variance is large, it is determined that the density change is drastic. If the variance is small, the density gradient is determined to be flat. It is determined that the density distribution is appropriate. However, the present invention is not limited to the above method. For example, the pixel density value may be differentiated with respect to the spatial axis, and the number of local maximum values and / or local minimum values may be used as a criterion.

Since the encoders 4a to 4c have the same basic configuration, they will be described in common with reference to FIG.

When the blocked image information 3 is input, the quantization level calculator 10 outputs the input image information 3
, The quantization level 11 is calculated. The number and value of these quantization levels are different for each encoder. For example, in the encoder 4b that performs the two-tone block approximate encoding, as shown in FIG. Q1, Q2 2
Individual. In the case of 4 tones and 16 tones, the quantization levels are Q1 to Q, as shown in FIGS.
4 (Q2 and Q3 are not shown), Q1 to Q16 (Q
2 to Q15 (not shown)).

More specifically, as shown in FIG. 3, the maximum quantization level Q1 and the minimum quantization level Q2, or Q4 or Q16 are determined, and the interval between them is divided at equal intervals to perform linear quantization. The maximum quantization level Q1 is an average value of the density values of n pixels in order from the maximum value max of the pixel density in the block. Similarly, the average value of n pixels from the minimum density value min is set as the minimum quantization level. Here, assuming that the number of pixels in the block is L ² and the number of gradations is 1, the parameter n is a value on the order of approximately n = L ² / l. However, how to determine the maximum quantization level and the minimum quantization level is not limited to the above method, and for example, the maximum density value max and the minimum density value min in the block may be used as they are. Further, the present invention is not limited to the case where linear quantization is performed at equal intervals between the maximum quantization level and the minimum quantization level.
X quantization (Joel Max, “Quantizin
g for Minimum Distortion
n ”, IRE TRANSACTION ON INF
ORMATION THEORY, March 196
0, pp7-12). That is, an optimal quantizer may be designed assuming the probability density function of the input image.

Next, the gradation information encoder 12 encodes the quantization level 11 output from the quantization level calculator 10. Specifically, as shown in FIG. 3, the average value of the maximum quantization level Q1 and the minimum quantization level Q2, Q4 or Q16 is La, and the difference between them is Ld, and these values are output. In this embodiment, assuming that the input image information 3 has 256 gradations per pixel and is represented by 8 bits per pixel in an unsigned binary representation, La and Ld shown in FIG. Alternatively, it is expressed by a similar number of gradations and expressed by a total of about 16 bits. However, the encoding method of the quantization level 12 is not limited to this, and the maximum quantization level and the minimum quantization level may be assigned as they are. When the quantization level calculator 15 performs nonlinear quantization, it is necessary to encode the way of changing the quantization step (difference between adjacent quantization levels).

The quantizer 13 has a quantization level 11
, The input image information 3 is quantized, and the result is output as resolution information 14. In this case, the quantization threshold is an average value of the corresponding quantization levels. For example, the threshold between the quantization levels Q1 and Q2 is (Q1 + Q
2) / 2. In this embodiment, since the quantization is performed to 2 gradations, 4 gradations, or 16 gradations, the quantized resolution information 14 is information of 1 bit, 2 bits, or 4 bits per pixel.

Next, the resolution information selector 15 removes visually redundant information from the resolution information 14 in accordance with the number of gradations quantized by the quantizer 13, and replaces the necessary resolution information 16. Output. In the present embodiment, FIG.
As shown in (b) and (c), resolution information of all 64 pixels (= 8 pixels × 8 pixels) is selected in the case of two gradation coding, and 32 pixels in the case of four gradation coding. (= 8 pixels × 8
The resolution information of (pixel / 2) is selected in a zigzag pattern, and in the case of 16 gradation coding, 16 pixels (= 8 pixels / 2 × 8 pixels /
The resolution information of 2) is selected in the form of an orthogonal grid. In the drawing, a hatched portion indicates a selected pixel. However, the method of selecting the resolution information is not limited to the above example. For example, in the case of two-tone coding, resolution information of 32 pixels may be selected in a staggered pattern. Here, it is important to reduce the allocation of resolution information when a large amount of gradation information is allocated to a certain block, and to increase the allocation of resolution information when a small amount of gradation information is allocated. is there.

Next, the resolution information encoder 17 encodes the selected resolution information 16. In the present embodiment, the selected resolution information 16 is directly expressed as an unsigned binary number without using coding for suppressing informational redundancy, and the coded image information 5a to 5c is output.

The selector 6 shown in FIG. 1 selects the coded image information 5a-5c output from the encoders 4a-4c according to the block decision result 8, and outputs it.

As described above, the image information 3 input for each block is encoded into a combination of gradation information and resolution information.
It is output as encoded image information 9.

Table 1 shows the relationship between the gradation information and the resolution in the block approximation encoding of each gradation.

[0038]

[Table 1]

As shown in Table 1, in this embodiment, constant information (80 bits) is always allocated to each block and encoded, regardless of the result of block determination. When 2-bit block selection information is added for each block in order to identify which encoder has been selected at the time of decoding, the encoding efficiency or compression ratio of this embodiment is 8 pixels × 8 pixels × 8.
Bits / (80 bits + 2 bits) = 6.244.

The gradation information encoder 12 and the resolution information encoder 17 shown in FIG. 2 do not use coding for suppressing informational redundancy, such as Huffman coding or arithmetic coding. Shall be.

Next, the decoding procedure of this embodiment will be briefly described. First, from the block selection information, which of the encoders 4a to 4c in FIG. 1 is selected at the time of encoding is identified, and the gradation information La and Ld and the resolution information shown in FIG. Is decrypted. In the present embodiment, since encoding such as Huffman encoding is not employed, binary numbers arranged in a predetermined order and bit length may be sequentially read. Next, using the gradation information La and Ld, a range from La + Ld / 2 to La−Ld / 2 is divided at equal intervals, and quantization levels Q1, Q2,... Are calculated. Image information is reproduced from the obtained quantization level and the resolution information. At this time, if the resolution information is selectively missing in the horizontal or vertical direction, for example, interpolation is performed using the already reproduced surrounding pixel density values to reproduce the density of the missing pixels. With the above procedure, the encoded image is decoded.

FIG. 5 shows another embodiment of the present invention. The difference from the embodiment shown in FIG. 1 is that the respective encoded image information 5a to 5c simultaneously encoded by the encoders 4a to 4c are
The selector 6 is supplied to the local decoder 18 and the selector 6 is controlled by the output of the local decoder 18. The local decoder 18 decodes the coded image information 5a to 5c, compares the coded image information 5a to 5c with the image information 3 before being coded by the block determiner 7, and according to the result 8, the selector 6 has the least distortion. Select and output encoded image information. That is, the block determiner 7
Functions as distortion calculating means. As a measure of the distortion, a mean square error may be used, or a cumulative value of the absolute value of the difference may be used.

In another embodiment of the present invention shown in FIG. 5, an image input for each block is a block whose density gradient is relatively flat, a block whose density changes drastically, or an intermediate between the two. There is no need to determine whether a block is a block. In order to visually determine whether the image information input for each block is coded by any of the plurality of encoders with the least image quality degradation, the present embodiment shown in FIG. Although the variance value of the inner pixel density and the differential coefficient on the spatial axis are used, it cannot be said that the optimum determination is always performed. On the other hand, in another embodiment shown in FIG. 5, an optimal determination is made on at least a distortion measure such as a mean square error or the accumulation of absolute values of differences. Therefore, it is possible to perform encoding with less deterioration in image quality.

Next, in still another embodiment of the present invention,
This will be described with reference to FIG. 5 is different from the embodiment shown in FIG. 5 in that each of the coded image information 5a to 5c simultaneously coded by the coder 4a to 4c is locally provided in correspondence with each of the coder 4a to 4c. Decoders 18a to 18
c, respectively, and decodes them.
The difference is that the decoded image blocks 19a to 19c from 8c are compared with the input image block 3 to select the coded image information with the least distortion. That is, each local decoder 18
The decoded image blocks 19a to 19c from a to 18c are supplied to a distortion amount measuring unit 20, and the decoded image blocks 19a to 19c for the input image unit 3 are supplied to the distortion amount measuring unit 20.
, And outputs mode information 21 for determining an encoder capable of obtaining a decoded image block having the least distortion, that is, the highest degree of approximation. Based on this mode information 21, the selector 6 performs encoding. The one with the least distortion is selected from the image information 5a to 5c. The selected BTC encoded data 22 is multiplexed with the mode information 21 in the multiplexing unit 23, and encoded data 24 for each input image block 3 is obtained.

Next, the operation will be described in detail.

The digital input image 1 is divided into blocks of m × n pixels by the blocking unit 2. A BTC encoder 4a is provided for each of the input image blocks 3 thus obtained.
４4c are coded in k ways from mode 1 to mode k.

The BTC encoding mode is, for example, k = 3
Mode 1: Combination of 1/2 subsample and 2-level quantization Mode 2: Combination of 1/4 subsample and 4-level quantization Mode 3: Combination of 1/8 subsample and 16-level quantization The combination of the sub-sample for approximating the resolution information and the quantization for approximating the gradation information is set in advance. Note that a subsample means thinning out an image. For example, a １ ／ subsample means thinning out pixels to reduce the number of pixels to １ ／.

Assuming that the sub-sample ratio of mode i (where 1 ≦ i ≦ k) is ss (i), the number of quantization levels is ql (i), and the number of pixel gradation levels in the input image block 3 is N , The compression ratio r is given by r = ss (i) × log ₂ (ql (i)) / log ₂ (N). Here, if a combination of the sub-sample ratio ss (i), the number of quantization levels, and ql (i) where r is constant is set in advance, the input image block 3 is always encoded with a constant code amount. Can be encoded by specifying a compression ratio in advance.

Next, the BTC encoded data 5a to 5c obtained by the BTC encoders 4a to 4c are supplied to k kinds of local decoders 18a to 18c, and k kinds of BTC decoded image blocks 19a to 19c are respectively supplied. From these, the BTC decoded image block closest to the input image block 3 is obtained by the distortion amount measuring unit 20.

Here, each pixel in the input image block is represented by s
_l (l = 1,2, ..., mn), BTC in mode i
Each pixel of the coded and locally decoded BTC decoded image block is represented by x _l (i) (l = 1, 2,.
n) and S = {s ₁ , s ₂ ,. . . , S _{m · n} ｝, X
(I) = {x ₁ (i), x ₂ (i),. . . , X _{m · n}
(I) If｝, an evaluation function d (S, X
(I)) is, d (S, X (i )) = Σ | s l -x l (i) |: L1-Norm or, d (S, X (i )) = Σ (s l -x l (I)) ² : given by L2-Norm. Hereinafter, this is referred to as a distortion amount for mode i. Note that L2-Norm indicates a Euclidean distance indicating a distance between two points on the multidimensional coordinate axis, and L1-Norm.
Indicates a Chebyshev distance that is an approximate solution of this distance.

The distortion amount measuring section 20 obtains the amount of distortion for each of the modes 1 to k, selects the mode with the least amount of distortion, and determines the BTC encoding mode for the input image block. When this operation is expressed, for all i, mode = min {d (S, X (i))} (1 ≦ i ≦ k), and since the total number of modes is k, k−1 times, An evaluation function d (S, X (i)) is calculated. Note that mode is a variable indicating the mode.

By the processing up to this point, the BTC encoding mode (mode information) 21 that minimizes the amount of distortion is obtained.
Finally, after selecting the BTC coded data 22 corresponding to the BTC coding mode 21 with the selector 6, the multiplexing unit 23 multiplexes the BTC coded data 21 and outputs the BTC coded data 24 for the final input image block. Get.

FIG. 7 is a diagram for explaining the principle of the BTC encoding operation. Generally, human visual characteristics have the following properties.

(A) Flat image area: High sensitivity to gradation and low sensitivity to resolution.

(B) Complex image area: High sensitivity to resolution and low sensitivity to gradation.

This property is known as Weber's law. Therefore, it is not necessary to save all data of the significant information of the original image of 256 gradations and 400 dpi depending on the characteristics of the image area. Therefore, in the encoding device shown in FIG. 6, a combined BT that approximates the number of gradations and the resolution is used.
A plurality of C encoding modes are set, and approximate encoding is performed in the mode closest to the input image among these modes.
For example, when encoding an image having a large change in density, the resolution is prioritized, and the encoding is performed by increasing the resolution and decreasing the number of gradations. Conversely, when encoding an image having a gradual change in density, encoding is performed by increasing the number of gradations while giving priority to gradation and lowering the resolution. In this way, the code length can be made constant. Fixed-length encoding corresponds to arranging the combinations of the BTC encoding modes on a straight line in the graph shown in FIG.

In each of the above embodiments, the number of bits allocated to the number of gradations and the number of bits allocated to the resolution are dynamically switched in accordance with the characteristics of the image, so that efficient compression processing is performed with a fixed number of bits. be able to.

FIG. 8 shows still another embodiment of the present invention in which the expandability and feasibility in improving the encoding performance are improved, and an optimal mode in terms of visual characteristics can be selected. This will be described with reference to FIG.

The input image 1 is transformed into m
After being converted into an input block 3 composed of × n pixels, it is supplied to an average value separator 25. The average value separator 25 calculates the average value of the input block 3 and outputs it as average value information 26, and also subtracts the average value from the input block 3 and outputs the result as an average value separation block 27. Mode discriminator 2
8 analyzes the statistical and spatial features of the average value separation block 27, and obtains approximate parameters, ie, mode information 29, in the approximate coding according to the analysis result. The adaptive approximation encoder 30 performs resolution approximation and gradation approximation on the average value separation block 27 according to the mode information 29, and outputs an approximation block 31. The details of the mode discriminator 28 and the adaptive approximation encoder 30 will be described later.

Average value information 2 from the average value separator 25
6. The mode information 29 from the mode discriminator 28 and the approximation block 31 from the adaptive approximation encoder 30
And multiplexed and output as code data 33.

The details of the mode discriminator 28 shown in FIG. 8 will be described with reference to FIG. The average value separation block 27 from the average value separator 25 is supplied to a waveform analyzer 34 that analyzes waveform information among statistical and spatial feature values.

The waveform analyzer 34 performs a pattern analysis of a set (set) of representative vectors having representative waveform information and an average value separation block 27 which is a block composed of m × n pixels.
A pattern matching unit 35 that performs matching, selects a representative vector having the closest waveform information, and outputs a vector index 36 representing the selected representative vector.
And a waveform mapping table 37 for obtaining candidate values of parameters related to resolution approximation from the vector index 36 and outputting the obtained values as waveform information 38.

The average value separating block 27 is also supplied to a gain analyzer 39 for analyzing gain information in a block composed of m × n pixels among the statistical and spatial features. The gain analyzer 39 calculates m
A variance calculator 40 which calculates and outputs a variance value 41 of the value of × n pixels, and a histogram counter 42 which counts the frequency distribution of the value of m × n pixels of the average value separation block 27 and outputs the result as histogram information 43 And a gain mapping table 44 for obtaining candidate values of parameters related to gradation approximation from the variance value 41 and the histogram information 43 and outputting the obtained values as gain information 45.

The output of the waveform analyzer 34 and the gain analyzer 39
Is supplied to the mode determiner 46 and the waveform information 38
And the gain information 45, the mode information 29 which is an approximate parameter is obtained.

Next, the adaptive approximation encoder 3 shown in FIG.
0 will be described in detail with reference to FIG. The average value separation block 27 is supplied to an adaptive sub-sampler 47 for performing sub-sampling for approximating the resolution.
Average value separation block 2 based on sub-sample pattern 49 supplied from sample pattern generator 48
Seven m × n pixels are sub-sampled. At the time of this sub-sampling, the sub-sample pattern generator 48 generates a sub-sample pattern 49 according to the parameters related to the resolution approximation in the mode information 29.
Control. The sub-sample block 50 sub-sampled by the adaptive sub-sampler 47 is supplied to the adaptive quantizer 51. The adaptive quantizer 51 performs tone approximation, that is, quantization, in accordance with parameters relating to tone approximation in the mode information 29, and outputs an approximate block 31.

In this quantization, first, a dynamic range is obtained from the maximum value and the minimum value of the pixel values in the average value separation block, and the pixel values in the average value separation block are normalized by this dynamic range. After that, quantization is performed in accordance with parameters related to gradation approximation. Further, the maximum value and the minimum value of the pixel values in the average value separation block are quantized with predetermined characteristics and then added to code data described later.
However, it is needless to say that the maximum value and the minimum value do not always need to be quantized.

Next, the operation of the embodiment shown in FIGS. 8 to 10 will be described.

As schematically shown in FIG. 11A, an input image 1 input by raster scanning is converted into m × n pixels by a blocker 2, as schematically shown in FIG. Scan-converted into an input block 3 consisting of Thereafter, the encoding process is performed independently for each input block 3 as a unit.

The input block 3 is supplied to an average value separator 25. The average value separator 25 calculates the values S _ij (i = 1, 2,..., M, j) of the m × n pixels constituting the input block 3.
= 1, 2,..., N), that is, average value information 26, and then m
The average value μ is subtracted from the pixel value of each of the × n pixels, and the value X _ij (i = 1, 2,.
, M, j = 1, 2,..., N) to calculate the average value separation block 27. Here, the relationship between S _ij , μ and X _ij is represented by the following equation.

[0070]

(Equation 3)

The mode discriminator 28 analyzes the statistical and spatial features of the average value separation block 27 and outputs mode information 29 as an approximate parameter according to the result.

As shown in FIG. 9, a waveform analyzer 34 which is a component of the mode discriminator 28 analyzes the waveform information and obtains a candidate value of a resolution approximation parameter. Also gain analyzer 3
9 analyzes the gain information and obtains a candidate value of the tone approximation parameter. Further, the mode determining unit 46 obtains an approximate parameter, that is, mode information 29, from the candidate values of the resolution approximation parameter and the candidate values of the gradation approximation parameter.

The waveform analyzer 34 calculates the average value separation block 2
The waveform information indicating the two-dimensional direction of the gradation change and the complexity of the gradation change are analyzed, and from the result, candidate values of resolution approximation parameters for approximating the resolution of the average value separation block 27 are obtained. .

The pattern provided in the waveform analyzer 34
The matching unit 35 performs waveform information analysis by pattern matching between a representative vector set having representative waveform information prepared in advance and an analysis target block (hereinafter referred to as an analysis block), that is, an average value separation block 27.

FIG. 12 schematically shows the contents of a representative vector set having representative waveform information prepared in advance.

First, samples of a training input block composed of representative images are normalized, and the normalized samples are defined as a group of samples distributed on a unit hypersphere. Next, a principal component, which is an axis for dividing the normalized sample into two, is obtained. The sample group on the unit hypersphere is applied to a well-known principal component analysis method, and the vector having the highest contribution ratio, that is, the first vector is used.
The image is divided into two using a hyperplane defined by the principal component vectors (see FIG. 12A). This hyperplane is a space including the representative vector 36. Then, it is checked which side of the principal component each vector of the sample is on the unit hypersphere. Next, FIGS. 12 (b1), (b2),
As shown in (c1) to (c4), the space is further divided into two using the hyperplane defined by the principal component vector in each of the divided spaces (the divided spaces # 1 and # 2) (the divided space #
3, # 4, divided spaces # 5, # 6). The above dividing process is repeated a predetermined number of times. When this is repeated n times, a representative vector set is obtained in the form of an n-stage binary tree. FIG.
1) to (d4) show centroid patterns corresponding to the respective divided spaces # 3 to # 6. FIG. 12 (d1) to (d4)
Indicates that directions indicating two-dimensional gradation changes are a horizontal direction, a vertical direction, an oblique left direction, and an oblique right direction, respectively. This corresponds to the direction v of the analysis block. FIG. 12 (e1) to FIG.
The sub-sampling pattern shown in (e4) is set, and the sub-sampling pattern is determined according to the selected center-of-gravity pattern. The sampling rate of this sub-sampling pattern corresponds to r.

The adaptive approximation encoder 30 finds the degree of approximation between the input block and the representative vector set by a binary tree search, and uses the history of the path up to the final stage as an index. One from the sampling pattern, and similarly one from the above sub-sampling rate corresponding to this index
Select each type. As a method of searching a binary tree,
For example, A. Buz et al ,: "Speech
coding based up vector
quantification, IEEE Trans.
Acoustic. Speech & Signal P
process, ASSP-28.5, pp. 526
A tree search method as disclosed in US Pat. In order to improve the efficiency of the search operation,
It is desirable to employ a method as proposed in the specification of the patent application “Image Signal Analysis System” filed on June 25, 1991 by the present applicant.

By the above-described waveform information analysis, the direction and complexity of the gradation change of the analysis block are obtained, and the vector index 36 is obtained from the index of the representative vector.

The operation for obtaining the degree of approximation is performed by pattern matching between the analysis block and the representative vector set.

The analysis block of m × n pixels is expressed as x = ｛x _i │
i = 1, 2,. . . , M × n} and a set of k representative vectors is represented by y = {y _i | _i = 1,
2,. . . , K}, the pattern matching can be defined by the following equation.

[0081] All i against _{d (x, y p) =} min {d (x, y i)} (i = 1,2, ...., k) where, d (x, y _i ) Is the distortion measure between x and y _i ,
It is defined by the square distortion or the like representing the Euclidean distance. p is the index of the representative vector, that is, the vector index 36, and indicates that the representative vector xp represented by _p has been selected as the representative vector having the waveform information closest to the analysis block. This vector index 36 is supplied to a waveform mapping table 37.

The waveform mapping table 37 outputs a candidate value of a resolution approximation parameter, that is, waveform information 38 from the vector index 36. The candidate values of the resolution approximation parameters are composed of the direction v (p) of the analysis block representing the two-dimensional direction of the gradation change, and the sub-sample rate r (p) representing the complexity of the gradation change.

The gain analyzer 39 shown in FIG.
The amplitude of the average value separation block 27 and the gain information indicating the frequency distribution of the pixel values are analyzed, and from the result, a candidate value of a gradation approximation parameter for approximating the gradation of the average value separation block 27 is obtained. The gain information analysis is performed by counting the variance and the histogram (cumulative frequency distribution) of the values of the m × n pixels forming the average value separation block 27.

The variance calculator 40 provided in the gain analyzer 39 calculates the variance 41 of the values of m × n pixels constituting the average value separation block 27. M × n with the average separated
The variance of a pixel is defined by the following equation.

[0085]

(Equation 4)

Alternatively,

[0087]

(Equation 5)

Here, the description will be made using the variance value σ.

The variance value σ is one or more threshold values, for example, high,
Compared to two thresholds that distinguish between medium and low, the variance is large,
It is determined whether the image belongs to medium or small, and the result of the determination is one of the feature amounts in the gradation direction.

As shown in FIG. 13, the histogram counter 42 counts the frequency by performing threshold processing on the average value separation block 27 based on the variance value σ. That is, the threshold is set to ± σ / a, and less than -σ / a, -σ / a or more and σ / a or less,
The frequency is counted at three places in a range larger than σ / a. Here, a is a real number of 1 or more, and in this embodiment, for example, a
= 3. The frequency values counted at the three locations are H ₋₁ ,
H ₀ and H ₁ . As shown in FIG. 13, H ₋₁ , H ₀ ,
From H _1, the histogram is unimodal distribution (FIG. (A) see)
It is determined whether the distribution is a bimodal distribution or a bimodal distribution (see FIG. 3B), and the result is obtained as histogram information 43. For example, H ₋₁ ≦
If H ₀ and H ₀ ≧ H ₁ , it is determined that the distribution is a unimodal distribution, and otherwise, it is a bimodal distribution.

This histogram information 43 is used to determine the type of image. That is, since the photographic image has a single-peak distribution and the character image has a double-peak distribution, the type of image can be determined from the distribution state.

Here, in the threshold processing by -σ / a, each pixel in the average value separation block 27 is divided by the variance in order to remove the influence of the difference in the dynamic range of each input block. This is equivalent to obtaining a histogram in a normalized state.

Next, the gain mapping table 44
From the variance value 41 and the histogram information 43, a candidate value of the tone approximation parameter, that is, gain information 45 is obtained. The candidate values of the tone approximation parameters include the tone approximation or quantization characteristic c and the number of levels l. Here, the quantization characteristic c indicates the type of distribution of the histogram.
For example, when the variance value 41 is large, the width of the density distribution is expected to be wide, so that the number of levels 1 is set to be large in order to enhance the gradation, but when the histogram information 43 indicates a bimodal distribution, That is, when the image is a character image, the number of gradations is reduced even if the variance 41 is large, so that the character is clearly encoded.

The mode determiner 46 obtains the approximate parameter, that is, the mode information 29 from the candidate value of the resolution approximation parameter and the candidate value of the gradation approximation parameter. When performing fixed-length encoding in which a fixed code amount is controlled for each m × n pixel, the sub-sample rate r among the candidate values of the resolution approximation parameter related to the code amount and the candidate value of the gradation approximate parameter are By manipulating the number of levels 1 in the above, the data amount of the approximation block 31 output from the adaptive approximation encoder 30 described later is made constant. Here, since the data amount of the approximation block 31 is proportional to p · log ₂ l, p · log _2
What is necessary is just to control the value of 2l constant. The controlled resolution approximation parameter and the gradation approximation parameter are output together as mode information 29. Further, if each of the candidate value of the resolution approximation parameter and the candidate value of the gradation approximation parameter is used as the mode information 29 as it is, the reproduction image quality can be kept constant.

The adaptive approximation encoder 30 shown in FIG.
According to the mode information 29, the approximation of the resolution and the approximation of the gradation are performed on the average value separation block 27.

The operation of the adaptive approximation encoder 30 will be described in detail with reference to FIG.

The sub-sample pattern generator 48
In the mode information 29, the resolution approximation parameter, that is,
A pattern 49 for sub-sampling the average value separation block 27 is obtained according to the direction v of the analysis block representing the two-dimensional direction of the gradation change and the sub-sample rate r representing the complexity of the gradation change. The pattern 49 is formed, for example, by dividing a block composed of m × n pixels by の み only in the vertical direction.
(See FIG. 12 (e1)), a pattern which is similarly thinned out only in the horizontal direction by half (see FIG. 12 (e2)), and a pattern which is similarly thinned out in the vertical and horizontal directions by half (FIG. 12 (e3)). ), (E4)) depending on the combination of the direction v and the rate r.

The adaptive sub-sampler 47 follows the sub-sampling pattern 49 and outputs the average value separation block 2
7 is sub-sampled.

The adaptive quantizer 51 quantizes the sub-sample block 50 in accordance with the tone approximation parameter, that is, the quantization characteristic c and the number of quantization levels 1 in the mode information 29. When the quantization characteristic c shows a single-peak distribution as shown in FIG. 13A, a quantizer optimal for a single-peak distribution (for example, a photographic image) is obtained. When a peak distribution is shown, a quantizer optimal for a bimodal distribution (for example, a character image) is selectively used. Here, as the quantizer, for example, the aforementioned MA
A non-linear quantizer such as an optimal X quantizer may be used.
That is, as disclosed in the above-mentioned MAX document, an optimal quantization characteristic can be designed for a known distribution, and therefore, a quantizer having the quantization characteristic may be used. The quantizer has a quantization characteristic c and a number of quantization levels l.
Each is prepared according to the combination of. For example, quantizers of 2, 4, 8, and 16 levels suitable for each of the single-peak distribution and the double-peak distribution are prepared, and are selected according to the characteristic c and the number of levels l.

The multiplexer 11 shown in FIG. 8 is obtained from the mode information 29 obtained from the mode discriminator 28, the average value information 26 obtained from the average value separator 25, and the adaptive approximation encoder 30. Multiplex the approximation block 31;
The code data 33 as shown in FIG. 14 is created. In addition,
As described above, when quantizing the dynamic range from the maximum and minimum values of the pixel values in the average value separation block and normalizing the pixel values in the average value separation block with the dynamic range, , The maximum value and the minimum value are quantized by predetermined characteristics and added to the code data 33.

According to the embodiment shown in FIGS. 8 to 10, the following effects can be obtained.

(1) Since the combination of the encoding modes can be freely changed by the mode determiner 46, the extensibility is excellent.

(2) Even if the types of coding modes are increased for improving the performance, the processing amount does not increase. (3) Criteria of Mode Discriminator 28, ie, Waveform Mapping Table 37, Gain Mapping Table 44,
By setting a criterion in consideration of visual characteristics in the mode determiner 46, it is possible to select an optimal mode in terms of visual characteristics.

Therefore, it is possible to realize an encoder suitable for encoding for a page memory that requires simple encoding processing.

Next, an embodiment in which the present invention is applied to color still picture coding will be described.

For color still picture coding, ISO
JPEG (Joint Photographic Ex), a joint organization of the International Organization for Standardization and the CCITT (International Telegraph and Telephone Consultative Committee)
ADCT (Adaptive Discrete Cosine)
Transform) system is recommended as an international standard system. The ADCT function is divided into an essential function for executing a basic application and an optional function for executing a wider range of applications. One of the optional functions is sequential reproduction coding. . In this sequential reproduction encoding, image information for one screen is not encoded at one time, but image information having low resolution and gradation is encoded first, and thereafter, image information having high resolution and gradation is encoded. Information is sequentially encoded.

In the case of performing the sequential reproduction encoding, an image component that cannot be reproduced by one encoding, that is, an encoding error is encoded again. In order to obtain an encoding error, a once-encoded image, that is, encoded data is decoded by a local decoder provided inside the encoder, and a difference from an image before encoding, that is, an original image is obtained. Therefore, it is necessary to hold the original image in the memory inside the encoder while the encoding and the local decoding are being performed once. This memory needs a large amount of memory to temporarily store at least the pixel values of at least one image. Therefore, in the embodiment described below, the memory capacity is reduced by encoding the image information and storing it in the memory.

FIG. 15 is a block diagram showing an embodiment of the present invention in which the memory capacity is reduced.

The input image 53 from the input image memory 52
Is ADCT via the resolution converter 54 and the subtractor 55
It is supplied to the encoder 56, and the ADC ×
Encoding is performed in units of n-pixel image blocks, and ADCT code data 57 is obtained. The ADCT encoder 56
Has a built-in blocking unit. Also, A
After the DCT code data 57 is decoded by the ADCT decoder 58, the DCT code data 57 is passed through an adder 59, for example, to an internal code that performs coding in image block units of m _l × n _l pixels (see FIG. 12C). The DCT local code image 61 is supplied to the encoder 60, is encoded again, and is stored in the page memory 62 as the internally encoded ADCT local code image 61. Note that the local encoder 60 calculates
It is assumed that a built-in block generator for obtaining an m _l × n _l block to be internally coded from an m × n pixel image block to be DCT coded. The internal coded ADCT local code data 63 from the page memory 62 is decoded by an internal decoder 64 that performs the reverse process of the internal coder 60 to obtain ADCT local decoded image information 65. The decoded image information 65 is supplied to the subtractor 55 via the resolution converter 66. The output of the resolution converter 66 is a delay circuit 6 for delaying one stage (one screen).
7 is supplied to the adder 60.

Next, the operation of the embodiment shown in FIG. 15 will be described. Note that the image information of the original image is sequentially encoded for each stage from low resolution to high resolution.

The input image memory 53 stores the image information of the original image, and the resolution converter 55 converts the resolution of the image information to a predetermined resolution. In the first stage, the image information is converted to a low resolution by the resolution converter 55, but is sequentially changed to a high resolution for each stage.

In the first stage, an input image 53 from an input image memory 52 is converted to a low resolution by a resolution converter 54 and supplied to one input terminal of a subtractor 55.
At this time, since there is no input to the other input terminal of the subtractor 55, the output of the resolution converter
And encoded, and output as coded data 57. The code data 57 in the first stage has a low resolution. By decoding the code data 57 of the low-resolution image information on the receiving side (not shown), low-resolution image information can be obtained. Code data 57 of this low-resolution image information
Is also supplied to the decoder 58, where it is decoded and
9 is supplied to one input terminal. At this time, the adder 5
9 has no input, the output of the decoder 58 is supplied as it is to the internal encoder 60, where it is coded into internal code data 61, and then the page memory 62
Is stored. The processing of the first stage ends here.

In the second stage, the input image 53 is converted to a medium resolution by the resolution converter 54,
5 is supplied to one input terminal. In the second stage, the local code data 6 from the page memory 62 is synchronized with the reading of the input image 53 from the input image memory 52.
3 is read out, decoded by an internal decoder 64 to obtain internal decoded image information 65, converted to a medium resolution by a resolution converter 66, and then supplied to the other input terminal of the subtractor 55. The signal is supplied to the adder 59 via the delay circuit 67. The image of the difference between the encoded image of the first stage and the image to be encoded in the second stage, that is, the encoding error image 68, which has been subtracted by the subtractor 55, is generated by the encoder 5
6 is encoded and output as code data 57. The code data 57 in the second stage has a medium resolution. If the receiving side decodes the code data 57 of the medium-resolution encoded error image and adds the previously decoded low-resolution image information to the image converted to the medium resolution, decoded image information of the medium resolution can be obtained. The code data 57 of the medium-resolution coded error image is also supplied to a decoder 58, where it is decoded and supplied to one input terminal of an adder 59.
Medium-resolution decoded image 69 added by adder 59
Are encoded by the inner encoder 60 and the encoded data 6
1 is stored in the page memory 62. So far the second
Stage processing ends.

Similarly, in the third stage, the input image 53 is supplied to one input terminal of the subtractor 55 through the resolution converter 54 at the same resolution. In the third stage, the code data 6 from the page memory 62 is synchronized with the reading of the input image 53 from the input image memory 52.
3 is read out, decoded by an internal decoder 64 to obtain internal code information 65, converted to a high resolution by a resolution converter 66, supplied to the other input terminal of the subtractor 55, and a delay circuit 67 Is supplied to the adder 59 via the. The coded error image 69 subtracted by the subtractor 55 is supplied to the coder 56, coded, and output as coded data 57. The code data in the third stage has a high resolution. If the receiving side decodes the code data 57 of the high-resolution encoded error image and adds the previously decoded medium-resolution image information to the high-resolution converted image, high-resolution decoded image information can be obtained.

In the embodiment shown in FIG. 15, the image is encoded by the local encoder 60 and then stored in the page memory 62, so that the memory capacity of the ADCT sequential reproduction encoder can be reduced. it can. At this time, ADC
The image block in the T encoder 56 and the local encoder 60
If the size of the image block is set so as to have a positive integer ratio relationship, the encoding / decoding operation of the local encoder 60 can be performed for each of the larger image blocks. it can. In this case, the local encoder 60 does not need to input and output image data with the ADCT encoder 56 after encoding and decoding all the images.

[0116]

As described above, according to the present invention, (1) even if the coding block size is increased in order to increase the coding efficiency, an encoder having two or more gradations is adaptively used. Therefore, the deterioration of the image quality is small.

(2) Since a plurality of encoders are used adaptively, the reproduction of the resolution of a character / line image or the like is visually important, and the reproduction of the gradation of a person / landscape image is visually important. For an image and an image having an intermediate property between the two, it is difficult to visually detect deterioration in image quality due to encoding. In addition, high coding efficiency can be achieved without using coding that suppresses informational redundancy such as Huffman coding.

(3) Since the coding efficiency is always kept constant for each block without employing coding for suppressing informational redundancy such as Huffman coding, the coded image information is stored in a memory having a certain capacity. For example, when the data is stored in a secondary storage device, or when the data is stored / reproduced by a secondary storage device having a constant transfer rate, it is not necessary to control the coding efficiency (code amount), and the capacity of the buffer memory can be reduced or omitted. Also, since only a certain part of the encoded image information encoded in a certain image unit can be accessed independently in a block size unit, an image is cut out as it is in the encoded image information,
Image editing such as transfer, move, and delete can be performed at high speed.

Further, for example, when the BTC encoding mode is determined without executing a plurality of block approximation encodings in advance, (4) there is no need to determine the encoding mode in advance. , The combination of encoding modes can be freely changed, and the scalability is high.

(5) Even when the types of coding modes are increased to improve performance, the processing amount required for coding / local decoding and distortion amount measurement is constant, so that the feasibility is high.

(6) Since the encoding mode is switched by using a feature amount that is important for visual characteristics, the subjective evaluation image quality at the same compression ratio can be greatly improved. Further, under the same image quality condition, the compression ratio can be improved. Further, the feature amount can be used for a secondary process after encoding such as an editing / printing process.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram of an image encoding device according to the present invention.

FIG. 2 is a schematic configuration diagram of an encoder in the image encoding device.

FIG. 3 is an explanatory diagram showing a state of quantization in a plurality of encoders.

FIG. 4 is an explanatory diagram showing a difference in resolution information between encoders.

FIG. 5 is a schematic configuration diagram of another embodiment of the present invention.

FIG. 6 is a schematic configuration diagram of still another embodiment of the present invention.

FIG. 7 is a graph for explaining the operation principle of the embodiment shown in FIG. 6;

FIG. 8 is a configuration diagram of an embodiment of the present invention using an adaptive approximation encoder.

FIG. 9 is a configuration diagram showing details of a mode discriminator used in the embodiment shown in FIG. 8;

FIG. 10 is a configuration diagram showing details of an adaptive approximation encoder used in the embodiment shown in FIG. 8;

FIG. 11 is an explanatory diagram showing scan conversion of an image.

FIG. 12 is an explanatory diagram showing the principle of waveform information analysis.

FIG. 13 is an explanatory diagram showing the principle of gain information analysis.

FIG. 14 is an explanatory diagram showing a configuration example of code data.

FIG. 15 is a configuration diagram showing an embodiment in which the present invention is applied to a sequential reproduction encoder.

FIG. 16 is an explanatory diagram showing an image divided into a plurality of blocks.

FIG. 17 is an explanatory diagram showing the structure of one block.

FIG. 18 is an explanatory diagram showing a reference density at the time of encoding.

FIG. 19 is an explanatory diagram showing an encoding state when the density change of an image is flat.

FIG. 20 is an explanatory diagram showing a state of encoding when the density change of an image is abrupt.

[Explanation of symbols]

1 input image, 2 blocker, 3 input block,
4a-4c BTC encoder, 5a-5c BTC encoded image information, 6 selector, 7 block determiner, 8 block determination result, 9 selected encoded image information, 10
Quantization level calculator, 11 quantization level, 12 gradation information encoder, 13 quantizer, 14 resolution information, 1
5 resolution information selector, 16 selected resolution information,
17 resolution information encoder, 18, 18a-18c local decoder, 19a-19c BTC decoded image block,
20 distortion amount measuring section, 21 mode information, 22 BTC encoded data, 23 multiplexing section, 24 encoded data, 2
5 mean value separator, 26 mean value information, 27 mean value separation block, 28 mode discriminator, 29 mode information, 3
0 adaptive approximation encoder, 31 approximation block, 32 multiplexer, 33 code data, 34 waveform analyzer, 35 pattern matching unit, 36 vector index, 37 waveform mapping table, 38 waveform information, 39 gain analyzer , 40 variance calculator, 41 variance value, 42 histogram counter, 43 histogram information, 44 gain mapping table, 45 gain information, 46 mode discriminator, 47 adaptive subsampler,
48 sub-sample pattern generator, 49 sub-sample pattern generator
Sample pattern, 50 sub-sample blocks, 51 adaptive quantizer, 52 input image memory, 53
Input image, 54, 66 resolution converter, 55 subtractor,
56 encoder, 57 code data, 58 decoder, 5
9 adder, 60 internal encoder, 61, 63 Locally encoded ADCT local code data, 62 page memory, 64 internal decoder, 65 ADCT local decoded image information, 67 delay circuit, 68 encoding error image , 69
Decrypted image

Continued on the front page (72) Inventor Kazuhiro Suzuki 2274 Hongo, Ebina-shi, Kanagawa Fuji Xerox Co., Ltd. Ebina Works (56) References JP-A-62-266924 (JP, A) (58) Fields investigated (Int .Cl. ⁷ , DB name) H04N 1/41-1/419 G06T 9/00 H03M 7/34

Claims (9)

(57) [Claims] 1. An image processing apparatus which samples an image and comprises m × x pixels
Blocking means for dividing into input blocks of n pixels (m and n are positive integers), block classification means for classifying the gradation and resolution of the input block, and code amount of gradation and resolution in the input block Comprises a plurality of block approximation coding means allocated to each input block so as to have a predetermined code amount, and adaptively switches the plurality of block approximation coding means according to the result of classification by the classification means. An image encoding apparatus for performing block approximation encoding by using a block. 2. A local decoding unit for decoding a plurality of pieces of encoded information obtained by simultaneously or sequentially encoding the input block by the plurality of block approximation encoding units, the block classification unit comprising: And a distortion calculating means for obtaining a degree of approximation to a plurality of local decoding blocks obtained from the means, and obtaining a classification result of the blocks from a minimum distortion among the plurality of local decoding blocks. Item 7. The image encoding device according to Item 1. 3. When decoding the encoding information of the image encoding device, if the resolution information is selectively missing in a predetermined direction, a corresponding pixel is interpolated from surrounding decoded pixels. 2. The image encoding apparatus according to claim 1, wherein the image is reproduced. 4. An image obtained by sampling an image, and
Blocking means for dividing into input blocks of n pixels (m and n are positive integers), average value calculating means for calculating an average value in the input block, and the average value calculating means from each pixel in the input block. Average value separating means for subtracting the obtained average value, analyzing means for analyzing feature amounts in the resolution direction and gradation direction of the average value separating block obtained by the average value separating means, and a result of the analyzing means. A mode determining means for independently determining a preset pixel thinning shape and a pixel thinning ratio in the plurality of average value separating blocks, and a gradation number of pixels in the average value separating block; and the mode determining means. Resolution approximating means for decimating pixels in the average value separation block according to the pixel decimating shape and the pixel decimating ratio determined by: Gradation approximation means for quantizing pixels in the average separation block with the number of gradations determined by the mode determination means, the average value from the average value calculation means, the determination result of the mode determination means, and the level code de multiplexes the output of the tone approximation means - a multiplexing means constituting the data, before
In the analyzing means, the characteristic of the input block in the resolution direction is
In the case of analyzing the collection amount, m × n pixels obtained in advance
(M and n are positive integers) or a positive integer ratio j thereof (j is a positive integer)
Multiple representative shape blocks consisting of pixels divided by an integer
The degree of approximation between each set of
Index of the representative block with the highest similarity
Or the degree of approximation for each block divided into j pieces
Input a set of indices of representative shape blocks with high
The first feature amount in the resolution direction of the block is at least
Of the index or the obtained j indexes
The ratio in which the pairs match each other is determined by the form of the average value separation block.
Parameter that indicates the complexity of the shape
An image coding apparatus , wherein the second feature amount in the image direction is used. 5. A pixel thinning shape, a pixel thinning ratio, and a floor thinning ratio in the input block by the mode determining means.
In the case where the tones are independently determined , first,
In the resolution direction of the input block determined by the analyzing means.
The average value separation prepared in advance from the first and second feature values
Pixel thinning shape and pixel thinning ratio candidates in the block
The input obtained by the analysis means independently of the input
It is used in advance from the first and second feature values in the gradation direction of the block.
Of the number of tones of pixels in the average value separation block
After obtaining the complement, the pixel thinning shape and the pixel thinning ratio
Pre-set among combinations of candidates and candidates for the number of gradations
Select a set with a constant compression ratio and select a set in the input block.
Determine pixel thinning shape, pixel thinning ratio, and number of gradations
In this way, a constant code amount is controlled for each input block.
Image encoding apparatus according to claim 4, characterized in that the control. 6. A pixel thinning shape, a pixel thinning ratio, and a floor thinning ratio in the input block by the mode determining means.
In the case where the tones are independently determined, first,
First and second resolutions of the input block determined by the analyzing means
And the second feature amount, the average value separation block prepared in advance.
Calculate pixel thinning shape and pixel thinning ratio candidates in lock
Independently, the input block determined by the analysis means is
Prepared in advance from the first and second feature values in the lock gradation direction
Of the number of gradations of the pixels in the average value separation block
Is calculated, and the pixel thinning shape and the pixel thinning ratio are evaluated.
A preset fixed value based on the combination of the complement and the number of gradation candidates
Select the set that has the playback image quality of
Pixel thinning shape, pixel thinning ratio, and number of gradations
By setting, a fixed playback image is
The image encoding apparatus according to claim 4, wherein the quality is controlled . 7. The method according to claim 7 , wherein the plurality of modes are determined by the mode determining unit.
In the case where the pixel thinning shape and the pixel thinning ratio in the average value separation block of
Orthogonal to the direction of the shape in each two-dimensional space
Set a set of pixel thinning shapes in the direction
A pixel in the average value separation block according to a second feature value;
5. The image encoding apparatus according to claim 4, wherein a set of thinning ratios is set . 8. The method according to claim 1, wherein the determination result of the mode determination means is
Pixels in the average value separation block by gradation approximation means
In the case of quantizing
The dynamic range is calculated from the maximum and minimum pixel values in the
And the pixel value in the average value separation block is
Normalized by the dynamic range and obtained by the analytical means
Prepared in advance according to the obtained second feature amount in the gradation direction.
One type is selected from a plurality of nonlinear quantization characteristics, and the mode is selected.
Normalized by the number of gradations determined by the
Quantizing the pixel values in the average value separation block
The maximum and minimum pixel values in the average value separation block
Quantizes the value with predetermined characteristics and adds it to the code data
7. The image encoding apparatus according to claim 5, wherein: 9. An image is sampled from low resolution to high resolution.
Resolution conversion means for sequentially converting to a predetermined resolution
And the image supplied from the first resolution conversion means.
Block of m × n pixels (m and n are positive integers)
Blocking means for dividing into blocks, and
Therefore, the first encoding of the input block
Block encoding means, and the first block encoding means
First local decoding of the input block coded in
Decoding means and the first local decoding means
The local decoding block is sent to the first block coding means.
It is the same as the size of the block to be cut, or an integer ratio.
The first block in the input block units is Uni set
Encoding in synchronization with the encoding operation in the encoding means.
2 block encoding means and encoded local decoded image
Is stored in units of images that are sequentially converted to a predetermined resolution.
Storage means and the locally decoded image stored in the storage means.
Second local decoding means for decoding, and the second local decoding means
Resolution of the locally decoded image from the first resolution conversion means
Second resolution conversion means for converting to a resolution corresponding to the degree
And the first resolution conversion means and the blocking means
Provided between the first and second resolution conversion means.
The station converted from the image by the second resolution conversion means
And a subtraction means for reducing a partial decoded image,
5. An image according to claim 4, wherein said second block coding means is used as said second block coding means.
An image encoding device using an image encoding device.